Interpretable Streetscape Quality Evaluation in Historic Districts: A Mask2Former–CRITIC–SHAP Framework

This dataset accompanies the manuscript "Interpretable streetscape quality evaluation in historic districts: A Mask2Former–CRITIC–SHAP framework. This dataset accompanies the manuscript submitted to Computers, Environment and Urban Systems." (submitted to Computers, Environment and Urban Systems). Contents: 1. streetscape_sample — 836 raw streetscape images (209 sampling points x 4 cardinal directions) covering 10 historic alleys in the Pingjiang Road Historic District, Suzhou, China. 2. preprocessed_sample — 836 preprocessed images. 3. segmentation_ade20k — 836 semantic segmentation masks from Mask2Former (Swin-L backbone, ADE20K 150 classes, PNG format). 4. ade20k_pixel_stats — Pixel count statistics (150 classes x 836 samples, CSV). 5. indicators — 15 streetscape indicators across 6 dimensions, CRITIC objective weights, and composite comfort scores. 6. shap_results — SHAP value matrix (836 x 15), feature importance summaries, and CRITIC-SHAP Spearman rank correlation (rho=0.757, p=0.001). 7. human_validation — Subjective visual quality ratings from 13 human raters. 8. ablation — Complete ablation experiment results. 9. statistical_tests — Pearson and Spearman correlation matrices, Mann-Whitney U test results. Code repository: https://github.com/3489229804-maker/mask2former-streetscape-assessment